Tumor Stromal Architecture Can Define the Intrinsic Tumor Response to VEGF-Targeted Therapy

Published OnlineFirst September 12, 2013; DOI: 10.1158/1078-0432.CCR-13-1637

Clinical Cancer Predictive Biomarkers and Personalized Medicine Research See related article by Rini, p. 6647

Tumor Stromal Architecture Can Deﬁne the Intrinsic Tumor Response to VEGF-Targeted Therapy

Neil R. Smith1, Dawn Baker1, Matthew Farren1, Aurelien Pommier1, Ruth Swann1, Xin Wang1, Sunita Mistry1, Karen McDaid1, Jane Kendrew1, Chris Womack1, Stephen R. Wedge1,2, and Simon T. Barry1

Abstract Purpose: The aim of the study was to investigate the vascular and stromal architecture of preclinical tumor models and patient tumor specimens from malignancies with known clinical outcomes to VEGFi treatment, to gain insight into potential determinants of intrinsic sensitivity and resistance. Experimental Design: The tumor stroma architecture of preclinical and clinical tumor samples were analyzed by staining for CD31 and a-smooth muscle actin (a-SMA). Tumor models representative of each phenotype were then tested for sensitivity to the VEGFR2-blocking antibody DC101. Results: Human tumor types with high response rates to VEGF inhibitors (e.g., renal cell carcinoma) have vessels distributed amongst the tumor cells (a "tumor vessel" phenotype, TV). In contrast, those malignancies where single-agent responses are lower, such as non–small cell lung cancer (NSCLC), display a complex morphology involving the encapsulation of tumor cells within stroma that also supports the majority of vessels (a "stromal vessel" phenotype). Only 1 of 31 tumor xenograft models displayed the stromal vessel phenotype. Tumor vessel models were sensitive to VEGFR2-blocking antibody DC101, whereas the stromal vessel models were exclusively refractory. The tumor vessel phenotype was also associated with a better Response Evaluation Criteria in Solid Tumors (RECIST) response to bevacizumab þ chemotherapy in metastatic colorectal cancer (CRC). Conclusion: The tumor stromal architecture can differentiate between human tumor types that respond to a VEGF signaling inhibitor as single-agent therapy. In addition to reconciling the clinical experience with these agents versus their broad activity in preclinical models, these ﬁndings may help to select solid tumor types with intrinsic sensitivity to a VEGFi or other vascular-directed therapies. Clin Cancer Res; 19(24); 1– 14. 2013 AACR.

Introduction cacy of these agents in a variety of preclinical tumor models in vivo The early promise of agents that target the VEGF signaling (3, 4). Given this discrepancy, further insight into axis has failed to translate widely in the clinic. Although potential determinants of VEGFi efficacy is warranted. VEGF signaling plays a pivotal role in angiogenesis, and The discordance between preclinical and clinical respo- some patients receive benefit, not all patients are responsive nses and the initial lack of markers identifying intrinsic to VEGF inhibitor (VEGFi) therapy, and those who do resistance to a VEGFi has been a significant focus of research. respond will eventually become refractory to treatment Additional studies in preclinical models have revealed a (1, 2). The expectation that the use of VEGFi therapy would number of putative mechanisms that may contribute to both transform the way in which many solid human tumors are intrinsic and acquired VEGFi resistance. One of the first managed clinically was based largely upon the broad effi- mechanisms proposed as mediating resistance to VEGF-A neutralization was the presence or recruitment of inflammatory infiltrate, specifically CD11b, GR-1–positive cells 1 Authors' Affiliations: Oncology Innovative Medicines, AstraZeneca, derived from the bone marrow, in response to upregulation Alderley Park, Macclesfield, Cheshire; and 2Northern Institute for Cancer Research, Newcastle University, Newcastle upon Tyne, United Kingdom of tumor-derived factors (5, 6). A complementary resistance mechanism is the expression of factors that drive alternative Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/). angiogenic pathways such as high expression of the potent Corresponding Authors: Neil R. Smith, Oncology iMed, AstraZeneca, angiogenic stimulus FGF-2 (7). Structural features of the Mereside, Alderley Park, Macclesfield, Cheshire SK10 4TG, United vessel have also been shown to influence the response to Kingdom. Phone: 440-1625233731; Fax: 440-1625510097; E-mail: VEGF signaling inhibitors, with mature vessels supported by [email protected]; and Simon T. Barry, Oncology iMed, AstraZeneca, Mereside, Alderley Park, Macclesfield, Cheshire SK10 pericytes or myofibroblast-like cells being less sensitive to 4TG, United Kingdom. Phone: 440-1625513350; Fax: 440-1625519749; VEGFi treatment. Furthermore, models of acquired resis- E-mail: [email protected] tance to VEGFi are characterized by the recruitment of doi: 10.1158/1078-0432.CCR-13-1637 pericytes, possibly through an elevated EGFR signaling 2013 American Association for Cancer Research. response (8), or platelet-derived growth factor (PDGF)-

www.aacrjournals.org OF1

Smith et al.

from experiments carried out as described (13) with licenses Translational Relevance issued under the UK Animals (Scientific Procedures) Act VEGF pathway inhibitors (VEGFi) are approved for the 1986. treatment of advanced cancer. Although these deliver some clinical benefit, questions remain as to which tumor DC101 tumor growth studies types are most sensitive, which patients receive greatest Calu-6 and Calu-3 human lung tumor xenografts estab- benefit, and why the broad preclinical activity seen with lished in nude and severe combined immunodeficient these agents did not translate to the clinic. Here, we (SCID) female mice, respectively. HT-29 and SW620 tumor differentiate human tumor types based on their stromal s were established in nude mice. Once tumors reached a architecture into tumor vessel or stromal vessel pheno- mean volume of about 0.2 to 0.3 cm3 (Calu-3), mice were types. The tumor vessel phenotype is associated with then intraperitoneally injected twice weekly with either 15 tumor types (renal and thyroid cancer) with better sin- mg/kg of DC101 (Cell Essentials Inc.) or an isotype control gle-agent clinical response to a VEGFi than the stromal antibody for the times indicated. DC101 is a monoclonal vessel phenotype (colorectal and non–small cell lung neutralizing antibody raised to murine VEGF receptor-2 cancers). The tumor vessel phenotype is also evident in (flk-1) that was chosen on the basis of its specificity as a many human tumor xenograft models commonly used VEGFi. Following treatment, tumors from each group were to evaluate efficacy, whereas a model displaying the excised and divided; one half being snap frozen in liquid stromal vessel phenotype was found to be refractory to nitrogen and stored at 80C until required, and the other treatment with the VEGFR2-blocking antibody DC101. fixed in neutral-buffered formalin for 24 hours. These data suggest an association between stromal architecture and intrinsic tumor sensitivity to VEGFi therapy. Histopathologic staining The following antibodies were used in IHC and immu- nofluorescent (IF) analyses: rabbit anti-mouse CD31 (AstraZeneca, CHG-CD31-P1; ref. 13); mouse anti- mediated signals (9–11). Determining the potential clinical human a-smooth muscle actin (a-SMA;Sigma,1A4); relevance of such resistance mechanisms requires a closer rabbit anti-human PDGFRb (Epitomics 1469-1); mouse examination of the preclinical models to determine which anti-human CD68 (Dako, M0876); mouse anti-human aspects are representative of human cancer. Neutrophil Elastase (Dako, M0752); rabbit anti-human We have previously studied the diversity in the angiogenic E-cadherin (Cell Signaling Technology, 3195); and response represented by a panel of preclinical tumor xeno- mouse anti-human vimentin (Dako, M0725), rat anti- graft models and found that while the human tumor cells mouse Gr-1 (BD Pharmingen, 550291), rat anti-mouse may show different expression of angiogenic genes, the host F4/80 (Serotec, MACP497), and mouse anti-human Ki67 response is similar between models (12). To build on this (Dako, M7240). Tissues were sectioned onto glass slides, study, we have profiled a number of different human dewaxed, and rehydrated. For both IHC and IF, all incu- tumors to examine the histologic relationship between bations were conducted at room temperature and TBS tumor and stroma, to determine whether these features are containing 0.05% Tween (TBST) used for washes. Antigen present within a panel of histologically diverse human retrieval was conducted in pH 6 retrieval buffer (S1699, tumor xenografts. We found that based on the tumor Dako) at 110C for 5 minutes in an RHS-1 microwave stromal architecture human disease could be broadly sub- vacuum processor (Milestone), then endogenous biotin divided into 2 phenotypes that have different sensitivity to (Vector, SP-2002, neutrophil elastase, E-cadherin, Gr-1 VEGFi therapy as a single agent. The significance of the and F4/80 only), peroxidase activity (3% hydrogen per- phenotype is explored in the context of both preclinical oxide for 10 minutes), and nonspecific binding sites modeling and clinical samples. (Dako, X0909) blocked. For single marker analyses, antibodies raised to CD31, Materials and Methods a-SMA, PDGFRb, CD68, neutrophil elastase, E-cadherin, Human and xenograft tumor tissues vimentin, Gr-1, F4/80, and Ki67 were diluted, 1:400, Formalin-fixed, paraffin-embedded (FFPE) human pri- 1:1,000, 1:2,000,1:200, 1:200, 1:100, 1:100, 1:1,000, mary cancer resection blocks, both whole and formatted 1:100, and 1:100, respectively, in antibody diluent (Dako, into tumor microarrays (TMA), were sourced under S0809) and applied to sections for 1 hour. Mouse Envision approved legal contract from commercial tissue suppliers, secondary (Dako, K4007) for a-SMA, vimentin, CD68, and Asterand, Cytomyx, and TriStar Technology Group and Ki67, rabbit Envision secondary (Dako, K4003) for CD31, Wales Cancer Bank. Appropriate consents, licensing, and biotinylated rabbit anti-mouse IgG (Dako, E0464) for neu- ethical approval were obtained for this research. The suit- trophil elastase, goat anti-rabbit IgG (Dako, E0432) for ability of each specimen for immunohistochemical (IHC) PDGFRb, and E-cadherin or rabbit anti-rat IgG (Dako, analyses was determined by pathology assessment of tissue E0488) for Gr-1 and F4/80 were added for 30 minutes. For morphology and preservation [hematoxylin and eosin neutrophil elastase, vimentin, Gr-1, and F4/80, Vectastain (H&E)] and the general extent of antigen preservation (pan Elite ABC solution (Vector, PK-6100), diluted as instructed p-Tyr immunostains). Tumor xenograft tissue was derived in kit, was added for 30 minutes. Sections were washed and

OF2 Clin Cancer Res; 19(24) December 15, 2013 Clinical Cancer Research

Tumor Stromal Phenotypes Deﬁne VEGF Sensitivity

developed in diaminobenzidine for 10 minutes (Dako, into tumor, stroma, and necrotic regions and used to K3466) and then counterstained with Carazzi’s hematoxy- determine the percentage of each compartment per tumor. lin. Appropriate no primary antibody and isotype controls This Genie classifier was combined with the appropriate were conducted for each antibody. Chromogenic CD31- Aperio image analysis algorithm to analyze biomarker a-SMA costain was conducted using the Envision G/2 parameters in stromal- or tumor-classified regions of doublestain System (Dako, K5361) following the manu- Calu-3 tumor s. Calu-3 stromal or tumor MVD, based on facturer’s recommendations with antigen retrieval and pri- CD31 immunostain, was determined using the microvessel mary antibody dilution as described above. CD31-a-SMA analysis algorithm. Stromal myofibroblast content (a-SMA co-IF was conducted as previously described (13). and PDGFRb) or macrophage (F4/80) and neutrophil (Gr- 1) infiltrate into the stroma or tumor was analyzed using the Pathology scoring and computer-assisted image color deconvolution algorithm (Leica Biosystems) to deter- analysis mine the percentage of positive pixels per total number of Chromogenic or fluorescent images were captured using pixels. Tumor proliferative index [number of Ki67-positive the 20 objective of either an Aperio image scan (Leica tumor cells/total number of tumor cells (hematoxylin-pos- Biosystems) or Pannoramic SCAN (3DHISTECH), respec- itive nuclei)] and tumor cell density [total number of tumor tively. Scanned images were scored by a trained pathologist cells (hematoxylin-positive nuclei)/tumor area] were deter- and 2 scientists using simple subjective reporting proce- mined using a nuclear algorithm (Leica Biosystems). To dures. Both human and xenograft tumors were scored for measure tumor nest size, the diameter of 44 to 220 distinct stromal vessel and tumor vessel (tumor vessel) phenotypes nests were measured for each a-SMA–immunostained using H&E or CD31-a-SMA stains. The tumor vessel phe- Calu-3 section (to highlight hematoxylin-positive tumor notype was defined as a tumor structure where vessels are nests) using the Aperio image viewer measuring tool (Leica embedded throughout the tumor cell mass and the stromal Biosystems). Biomarker data were analyzed using the Stu- vessel phenotype classed as tumor cell nests surrounded by dent 2-tailed t test to determine statistical significance well-developed stromal structures which contain the major- between treatment groups. ity of the vessels. A tumor was classified as either tumor vessel or stromal vessel based on the predominant phenotype TaqMan fluidigm gene expression profiling (cutoff of >60% tumor area), whereas a tumor composed RNA was isolated from 30 to 50 mg frozen tumor using an of 40% to 60% of both phenotypes was scored as interme- RNeasy Lipid Tissue Mini Kit (QIAGEN, 74104), according diate. These criteria were prospectively defined by an expert to manufacturer’s protocol. On-column DNase digestion pathologist to clearly differentiate between tumors that were was conducted using the RNase-free DNase Kit (QIAGEN, predominantly tumor vessel (>60%) or stromal vessel 79254). RNA concentration was measured using the Nano- (>60%) and those that were intermediate (40%–60%) and Drop ND1000 (Thermo Fisher Scientific). Human- and difficult to classify as one or the other phenotype. On the mouse-specific assays were designed and supplied by basis of CD68 marker immunostaining, macrophage infil- Applied Biosystems, whereas eukaryotic 18S rRNA was used trate into the tumor compartment was classified as negative, as the endogenous control (Supplementary Table S4). Total þ low (>0but1% CD68 cells: tumor cells) or medium-high RNA (50 ng) was converted into cDNA using the High þ (>1% CD68 cells: tumor cells). Tumor epithelial-to-mes- Capacity cDNA Reverse Transcription Kit (Applied Biosys- enchymal transition (EMT) status was scored using epithelial tems, 4368814) in final volume of 20 mL, according to the (E-cadherin) and mesenchymal (vimentin) markers to manufacturer’s instruction. cDNA (1.25 ml) was pre-ampli- þ define epithelial (E-cadherin vimentin ), mesenchymal fied using a pool of TaqMan primers at a final dilution of 1 þ (E-cadherin vimentin ), or intermediate phenotypes (E- in 100 and Pre-amplification Master Mix (Applied Biosys- þ þ cadherin vimentin ). tems, 4391128) in a final volume of 5 mL. Samples were Computer-assisted image analysis was conducted on diluted 1 in 5 with 1 TE and stored at 20C. Sample and digitally acquired chromogenic images. For images of assay preparation for 48.48 dynamic arrays was conducted human tumors immunostained for CD31-a-SMA, the according to the manufacturer’s instruction (Fluidigm). tumor compartment, including associated stroma, was Data were collected and analyzed using the Fluidigm selected by hand using the Aperio Image Viewer Software Real-Time PCR Analysis 2.1.1 software (v.2.1.3). Species- (Leica Biosystems). Downstream image analyses of the specific normalization of the expression data to 18S rRNA annotated areas were conducted using Aperio image anal- was conducted as reported (12). Gene expression values ysis software (Leica Biosystems). Microvessel density (MVD, were calculated using the comparative CT (DCT) method as number of vessels per mm2 viable tumor), using CD31 as a previously described in User Bulletin #2 ABI PRISM 7700 marker, was determined using the Aperio microvessel anal- Sequence Detection System 10/2001, using the corrected ysis algorithm (Leica Biosystems). Myofibroblast levels 18S rRNA Ct values for normalization of the tumor tran- based on a-SMA were determined using the Aperio color script and the original values for the stroma. To determine deconvolution algorithm to measure percentage of a-SMA– genes altered by DC101 treatment student and differentially positive pixels per total number of pixels. Genie (Leica expressed between Calu-3 versus the others models, t tests Biosystems), a pattern recognition software tool, was and fold change were calculated with significantly altered trained to segment stained images of the Calu-3 xenograft genes identified by having a P < 0.05 and a fold change >1.5.

www.aacrjournals.org Clin Cancer Res; 19(24) December 15, 2013 OF3

Smith et al.

Results (HNSCC), non–small cell lung carcinoma (NSCLC), ovar- Tumor stromal architecture deﬁnes human tumor ian (OvC), thyroid (ThC) subtypes, colorectal (CRC), pros- types that respond to VEGF signaling inhibitors as tate (PC), and breast (BC) cancers (Fig. 1; Supplementary monotherapy Table S2). For the purposes of this analysis, we have viewed To assess the tumor stromal architecture of different solid responsive disease as high incidence of tumor response by human malignances, we used a number of tumor TMAs Response Evaluation Criteria in Solid Tumors (RECIST) comprising renal cell carcinoma (RCC), glioblastoma mul- assessment (tumor shrinkage) and/or a large increase in tiforme (GBM), head and neck squamous cell carcinoma progression free survival on long-term therapy reported

A B

TV SV

RCC CRC

Figure 1. Two dominant phenotypes CD31 CD31 α αSMA based on tumor stromal SMA 100 μμm 200 μm μ 200 μm DAPI 100 M DAPI 100 m architecture stratify tumor types. HCC NSCLC Tumor vessel phenotype (A; a tumor structure where vessels are 100μM embedded throughout the tumor cell mass) and stromal vessel phenotype (B; a tumor structure dominated by a pattern of tumor cell

CD31 CD31 nests surrounded by well- α αSMA SMA 100 μm 200 μm μ 200 μm DAPI DAPI 100 m developed stromal structures which ThC PC contain the majority of the vessels). Also shown are IF-stained FFPE human tumors [CD31-AF488 (green), a-SMA -AF555 (red), and DAPI counterstain (blue)], indicative of tumor vessel (RCC, GBM, HCC, ThC, OvC, and HNSCC) and stromal CD31 CD31 αSMA μ αSMA vessel (CRC, NSCLC, PC, and BC). 100100 mμM 200 μm 100 μm 200 μm DAPI DAPI C, percentage ratios of tumor vessel: stromal vessel phenotypes C for RCC, GBM, OvC, HCC, ThC, = 50 = 56 = 36 = 30 = 37 = 49 = 26 = 61 = 49 = 44 HNSCC, CRC, NSCLC, PC, and BC. 100% n n n n n n n n n n Phenotypes were scored from a 90% TVTV CD31- -SMA or H&E-stained TMAs SV (1 TMA per tumor type, number of 80% SV patient samples per tumor type are 70% shown).

60%

50%

40%

Phenotypes (% ratio) Phenotypes (% 30%

20%

10%

0% PC BC ThC OvC HCC CRC RCC GBM NSCLC HNSCC

TV SV

OF4 Clin Cancer Res; 19(24) December 15, 2013 Clinical Cancer Research

Tumor Stromal Phenotypes Deﬁne VEGF Sensitivity

with different therapies targeting VEGF signaling. RCC (14) lesions revealed that the tumor vascular density in the and GBM (15, 16) are diseases we have taken as represen- combined tumor and stromal compartments was higher tative of cancers known to respond to VEGFi as a single- in tumor vessel (mean SEM ¼ 105.3 21.3) than in agent therapy (Supplementary Table S1). CRC (17) and stromal vessel (47.7 7.5; Fig. 2C and Supplementary Fig. NSCLC (18) are diseases where VEGF inhibitors show little S3b). a-SMA analysis showed myofibroblasts were greater single-agent responses and hence are trialed in combination in stromal vessel (mean SEM ¼ 10.0 1.8) than in tumor with cytotoxic therapy. Tumors were screened by IF to vessel (1.4 0.3) phenotype tumors (Fig. 2C and Supple- visualize the relative distribution of myofibroblast-like cells mentary Fig. S3b) tumors. EMT status was determined by (a-SMA), blood vessels (CD31), and tumor cells [40,6- immunostaining serial sections for E-cadherin and vimen- diamidino-2-phenylindole (DAPI)]. This revealed 2 dom- tin (Fig. 2D and Supplementary Fig. S3b). Tumor vessel inant morphologies: (i) a "tumor vessel" phenotype which tumors were generally more mesenchymal (E-cadherin / þ þ describes a tumor structure with vessels embedded through- vimentin ) than stromal vessel tumors (E-cadherin / out the tumor cell mass represented in RCCs, GBMs, and vimentin ). Qualitative histologic assessment of CD68- hepatocellular carcinoma (HCC; Fig. 1A, Supplementary immunostained tumors from the panel showed that the Fig. S1 and Supplementary Table S2); (ii) a "stromal- tumor vessel phenotype has higher levels of macrophage vessel" phenotype in which the tumor structure is dom- tumor infiltrate (Supplementary Fig. S3b and S3c). inated by a pattern of tumor cell nests surrounded by well- Neuroendocrine pancreatic cancer has recently been developed stromal structures containing the majority of shown to respond to the VEGFi sunitinib (23), whereas the vessels (Fig. 1B), as evident in CRCs, NSCLCs, and PC pancreatic adenocarcinoma has proven comparatively resis- (Fig. 1B, Supplementary Fig. S1 and Supplementary Table tant to such approaches (24–26). Comparing available S2). Using this classification, cores from multiple tumor samples of neuroendocrine pancreatic cancer with pancre- types were scored for either a tumor vessel or stromal atic adenocarcinoma revealed them to represent tumor vessel phenotype (Fig. 1C). The tumor vessel phenotype vessel and stromal vessel phenotypes, respectively (Supple- was observed in RCCs, GBMs, OvC, HCC, ThC, and head mentary Fig. S3a), consistent with our hypothesis on VEGFi and neck squamous cell carcinoma (HNSCC). The stro- sensitivity. mal vessel phenotype was dominant in CRC, NSCLC, PCs, and BCs. The tumor vessel phenotype is common to tumor As the phenotype ratios for each tumor type were derived xenograft models initially from TMA cores, full tumor sections were also To evaluate the relevance of the 2 phenotypes in preclin- classified for each phenotype to ensure the results were ical models, a panel of human tumor xenografts grown representative of larger diagnostic tumor samples [RCCs, subcutaneously and representing a broad range of different HCCs and PC (n ¼ 10); GBM, ThC, CRC, and BC (n ¼ 9); tumor types of origin [i.e., lung (8), colon (5), breast (4), and NSCLC (n ¼ 19); Fig. 2A, Supplementary Fig. S2 and prostate (3), brain (2), stomach (1), ovary (1), pancreas (1), Supplementary Table S3]. Where sections exhibit regions of skin (1), pharynx (1), uterus (1), vulva (1), and blood (1) both phenotypes, the predominant phenotype (>60%) was tumor and fibrosarcoma (1) cell lines] were classified for the scored (Supplementary Fig. S2a). The frequency of the 2 phenotypes based on the approach used for human phenotypes representing diagnostic specimens agreed with tumor s (Fig. 3). The majority (30 of 31) exhibited a the original TMA dataset (Supplementary Fig. S2b and S2c). tumor vessel phenotype (Fig. 3A and B). Only Calu-3 These data establish that the tumor stromal architecture (lung adenocarcinoma) had a stromal vessel phenotype of the tumor can be used to cluster human tumor types (Fig. 3A and C). These data indicate that the tumor vessel into 2 groups. Significantly, those clinical tumor types phenotype is common to subcutaneously grown tumor associated with response to a VEGFi alone [RCC (14), xenograft models. ThC (19, 20), GBN (15, 16)] were dominated by a tumor vessel phenotype. In contrast, tumors where a VEGFi has Intrinsic gene expression differences between stromal shown modest activity in combination [CRC (17), PC vessel and tumor vessel xenograft tumors (21), NSCLC (18), BC (22)] were dominated by the Gene expression analysis of human transcripts was used stromal vessel phenotype. to investigate the relationship between Calu-3, Calu-6, Additional biomarker analyses were conducted on the SW620, and HT-29 tumor s. The difference in expression diagnostic tumor panel (n ¼ 32 tumor vessel, n ¼ 35 stromal of a range of 180 genes associated with angiogenesis, vessel) to further examine the properties of the two phe- inflammation, and invasive growth was determined (Fig. notypes (Fig. 2B–D). Although immature and mature peri- 3D and E). This gene set differentiated Calu-3 from the other þ cyte covered vessels (CD31 endothelial cells with tightly models; genes showing the highest differential expression þ associated a-SMA cells) were associated with both phe- in Calu-3 were MMP7, LPAR3, CT55, IL1RN, and SPP1 notypes, tumor-embedded vasculature of the tumor vessel (osteopontin). Compared to other models analyzed, Calu-3 phenotype tended to be pericyte-free, whereas stromal xenograft tumor s were enriched in genes associated with vessels in the stromal vessel phenotype were often pericyte recruitment of the stromal cells, in particular fibroblast þ covered or associated with a-SMA fibroblasts (Fig. 2B). growth factor (FGF) and PDGF ligands (Fig. 3D). To vali- Computer-assisted image analysis of CD31 immunostained date that Calu-3 tumors showed a different expression

www.aacrjournals.org Clin Cancer Res; 19(24) December 15, 2013 OF5

Smith et al.

A B RCC CRC = 10 = 9 = 10 = 9 = 9 = 19 = 10 = 9 100% n n n n n n n n

90%

80%

70% 60%

50% CD31 CD31 40% α SMA 50 μm5αSMA 0 μm DAPI DAPI 30%

Phenotype (% ratio) Phenotype (% ThC NSCLC 20%

10%

0% BC PC ThC RCC CRC HCC GBM NSCLC

TV SV

TV 50 μm 50 μm Intermediate SV

CDTV Intermediate 40 500 SV = 41 = 38 100% n n 450 35 ) Mesenchymal

2 90% 400 30 80% Indeterminable Epithelial 350 70% 25 300 60% 20 50% 250 MVD 40% 200 15 30% 150 10 20% EMT status (% EMT status ratio) 100 α -SMA (%) positivity 10%

(number vessels per mm 5 0% 50 SV TV 0 0

PC BC Phenotype ThC PC BC RCC HCC CRC GBM ThC RCC HCC CRC GBM NSCLC (SCC) NSCLC (SCC) NSCLC (Adeno) NSCLC (Adeno)

Figure 2. Biomarker analyses of the 2 phenotypes in whole tumor samples. A, percentage ratios of tumor vessel:stromal vessel:intermediate phenotypes for RCC, GBM, HCC, ThC, HNSCC, CRC, PC, and BC. Phenotypes were scored from CD31-a-SMA costained whole FFPE tumor sections (number of patient samples per tumor type are shown). B, pericyte coverage of vessels associated with tumor vessel (RCC and ThC) and stromal vessel (CRC and NSCLC) þ phenotypes, white or gray arrows indicate naked and pericyte covered vessels, respectively. Samples were analyzed for: (C) MVD (number of CD31 vessels þ per mm2) and myoﬁbroblast content (percentage number of a-SMA pixels/total number pixels) and (D) EMT status [percentage ratio epithelial þ þ þ þ (E-cadherin vimentin ):intermediate (E-cadherin vimentin ):mesenchymal (E-cadherin vimentin )].

proﬁle, they were also compared to a broader panel of The tumor vessel phenotype has greater sensitivity to tumor xenograft models, again the same genes were over- VEGFR2 inhibitor, DC101, than the stromal vessel expressed relative to other models. This expression analysis phenotype, in vivo supports the conclusion that Calu-3 tumor cells exhibit high To determine whether the tumor stromal architecture expression of a number of transcripts that may contribute to could inﬂuence tumor response to VEGFi, we used the their phenotype. murine VEGFR2-blocking antibody, DC101 to assess the

OF6 Clin Cancer Res; 19(24) December 15, 2013 Clinical Cancer Research

Tumor Stromal Phenotypes Deﬁne VEGF Sensitivity

Tumour xenograft models (n = 31) A TV SV PC9 PC3 LoVo A375 A431 A549 HT29 FaDu Calu3 Calu6 HL-60 HPAC A2780 Du145 MCF-7 HX147 SW620 U87MG NCIN87 HT1080 ZR-75-1 BT474cl HCT116 Colo205 MES-SA U118MG NCIH460 NCIH526 NCI H1975 CWR22Rv1 MDAMB231

B TV C SV

200 μm 200 μm

D E

ddCT Calu-6 HT29 SW620 Calu-3 8.2 MMP7 7.5 LPAR3 7.4 CTSS 7.2 lL1RN SPP1 Calu-6 HT29 SW620 Calu-3

6.8 ddCT CCL5 6.8 −9.1 PGF FGF2 6.5 −8.9 GAS6 6.5 lL1R1 −6.6 MMP2 6.0 PDGFA −6.3 5.9 PDGFD FGF19 − 5.6 CDH1 6.1 DLL4 − 5.2 CDH3 6.0 ENG 5.2 CXCL1 −5.7 S100A4 4.9 CX3CL1 −5.2 ACVR1B CXCR7 4.7 −4.5 HEY1 PLAT 4.6 −4.2 MERTK 4.6 NOTCH3 −3.9 KISS1 4.3 PLAU −3.7 4.0 SERPINB5 MMP13 − 3.9 PDGFC 3.2 TYRO3 − 3.7 THBS1 3.1 ACVR1C 3.7 TIMP3 −3.0 ENPP2 3.6 MGP −3.0 F2R IL8 3.5 −2.9 SERPINE1 MST1R 3.4 −2.7 ANGPT1 3.3 LPAR1 −2.6 LPAR1 3.3 CCL22 −2.6 EFNB2 3.0 BMP2 −2.6 2.5 BMPR1B NOTCH1 − 2.5 CTSL1 2.5 BMPR1A 2.4 CTSD −2.4 FGF1 2.2 ID4 −2.3 NRP2 SGK2 2.2 −2.0 SPHK2 IL6R 1.9 −1.9 FGF18 1.9 EDG3 −1.7 CST3 1.9 IL6R −1.7 1.8 CTSL2 ITGA5 −1.7 1.8 IL1B FGF17 − 1.7 IL1A 1.6 BMP4

Figure 3. Prevalence and intrinsic gene expression differences between the 2 phenotypes in tumor xenograft models. A, classification of 31 subcutaneously implanted human tumor xenograft models for tumor vessel and stromal vessel phenotypes. Data were derived from the scoring of CD31-a-SMA– immunostained TMAs (4 FFPE tumors per model, 3 cores per tumor). B and C, diagrams and fluorescent immunostained [CD31-AF488 (green), a-SMA -AF555 (red), and DAPI counterstain (blue)] images representing preclinical xenograft. B, tumor vessel (Calu-6—low myofibroblast content) and (C) stromal vessel (Calu-3). D and E, transcript profiling of tumor vessel (Calu-6, HT-29, and SW620) and stromal vessel (Calu-3) tumor xenograft models. Genes with high (D) and low (E) expression levels in Calu-3 compared to Calu-6, HT-29 and SW620 are presented. ddCT ¼ [(average dCT Calu-3) (average dCT (Calu-6 þ SW620 þ HT-29)].

www.aacrjournals.org Clin Cancer Res; 19(24) December 15, 2013 OF7

Smith et al.

Calu-6 HT-29 SW-620 A 1.6 1.0 2.0 ) 1.4 0.9 ) 1.2 0.8 1.5 1.0 0.7 ± SEM ± SEM

± SEM) 3 3 TV 3 0.8 0.6 1.0 0.6 0.5 0.4 0.4 0.5 0.2 0.3

0.0 0.2 volume(cm 0.0 volume (cm Geometric mean tumour tumour mean Geometric Geometric mean tumour mean Geometric volume (cm volume Geometric meantumour 2824201612840 840 1612 840 201612 Days of treatment Days of treatment Days of treatment Figure 4. The tumor vessel Calu-3 (1) Calu-3 (2) phenotype is more sensitive to 1.8 1.8 VEGFR2 signaling inhibitor,

B ) ) 1.6 1.6 1.4 DC101, than the stromal vessel 1.4 in vivo 1.2 1.2 phenotype .AandB, ± SEM

± SEM

3 1.0

3 1.0 geometric mean tumor volume SV 0.8 0.8 growth curves (cm3) for DC101 (15 0.6 0.6 0.4 0.4 mg/kg twice weekly; red) and IgG1 0.2 0.2 (control; blue) treated arms of (A) volume (cm

volume (cm 0.0 0.0 Geometric mean tumour 0 5040302010 Geometric mean tumour 50 353025201510 Calu-6, HT-29, SW620 and (B) Days of treatment Days of treatment Calu-3 (growth curve 1, growth curve 2) tumor xenograft studies. Calu-6 Calu-3 (1) Calu-3 (2) Geometric mean SEM are shown C for growth curves. C, MVD (number 300 300 þ 2 ) of CD31 vessels per mm )ofIgG1 2

50 ) 2

) control and DC101-treated Calu-6

250 2 250 40 P = 0.004 P = 0.03 and Calu-3 tumor studies (growth P = 0.0001 200 ﬁ 30 200 curves 1 and 2). D, signi cant gene TV SV 150 changes induced in the tumor MVD MVD 20 150 (human) and stromal (stromal) 100 MVD vascular modulation and invasion 10 100 50 transcript proﬁle on DC101

(number vessels per mm per (number vessels 0

(number vessels per mm per (number vessels treatment of Calu-6 and Calu-3 0 50 (number vessels per mm per (number vessels (growth curve 1) xenograft tumors. Control Control 0 Tumor transcripts are normalized DC101 treated DC101 treated to human 18S, whereas stromal Control transcripts are normalized to total Calu-6 Calu-3 (1) ﬁ D DC101 treated 18S. Only signi cant gene changes are included (P > 0.05 and fold Human Human change > 1.5), Each column is an Gene fold P value Gene fold P value change Control DC101 treated change Control DC101 treated individual tumor and each color CDH8 5.17 0.005 PLAT 1.77 0.032 gradient is scaled independently SGK2 2.57 0.029 VEGFA 1.61 0.009 LPAR1 2.38 0.016 BMP2 1.53 0.037 within each gene. GAS6 1.70 0.028 MST1 1.51 0.036 LPAR1 1.68 0.034 FGF2 −1.69 0.033 SPP1 −1.98 0.014 CSTA −4.06 0.010 Mouse Gene fold P value Mouse change Gene fold P value Notch3 −1.50 0.034 change Mmp9 −1.52 0.020 Cxcl2 7.01 0.006 Cdh3 −1.65 0.034 Mmp9 2.55 0.027 Flt1 −1.70 0.005 ll1b 2.11 0.009 Mmp13 −1.72 0.031 ltgam 1.95 0.010 Gpr116 −1.76 0.030 − 0.005 Tie1 1.63 Tie1 −1.85 0.021 − Flt4 1.64 0.004 −1.88 0.042 − Efnb2 Pdgfb 1.66 0.003 −1.91 0.001 − Flt4 Gpr116 1.80 0.008 Angpt2 −2.03 0.021 − Efnb2 1.92 0.002 Dll4 −2.77 0.000 −2.02 Kdr 0.001 Cdh8 −4.03 0.009 Dll4 −3.32 0.001

effects of pruning neovasculature on tumor growth in tumor The gross vascular response to DC101 is similar vessel (Calu-6, HT-29 and SW620) compared to stromal between tumor vessel and stromal vessel tumor vessel (Calu-3) tumor xenograft models. DC101 was dosed xenografts intraperitoneally at 15 mg/kg twice weekly. Calu-6, HT-29, To compare the total vascular response of tumor vessel and SW620 xenografts exhibited the classic tumor growth and stromal vessel models to DC101, MVD analysis was response to VEGF signaling inhibitors (Fig. 4A) seen in conducted on CD31-immunostained control and DC101- multiple models (27–29). In contrast, Calu-3 tumors exhib- treated samples from the Calu-6 and Calu-3 (replicates 1 ited a poor response, with DC101 having no effect on tumor and 2) growth studies. The effect of DC101 on MVD growth (experiment 1) or inducing a small initial reduction reduction was similar between Calu-6 and Calu-3 models in growth (0–3 days, experiment 2) followed by a rapid (Fig. 4C). Furthermore, species-speciﬁc reverse transcrip- return to a growth rate similar to the controls (Fig. 4B). tion quantitative polymerase chain reaction (RT-qPCR)

OF8 Clin Cancer Res; 19(24) December 15, 2013 Clinical Cancer Research

Tumor Stromal Phenotypes Deﬁne VEGF Sensitivity

using genes associated with angiogenesis and tumor cell tumor vessel and stromal vessel phenotypes by 2 observers. invasion (Fig. 4D) revealed that human tumor gene changes Forty-two patient samples were classified as the stromal were unique to each tumor xenograft model, but a range vessel phenotype and 14 as the tumor vessel phenotype of murine genes associated with endothelial cells, were (Fig. 6). RECIST response information for FOLFIRI and reduced in both. Therefore, while there is a clear vascular bevacizumab (Avastin) treatment as first- or second-line response to DC101 in both models this does not translate to therapy post-surgery was available for each patient. Fisher growth inhibition in the Calu-3 xenograft. exact test was used to determine the significant associations between the phenotype and best RECIST response catego- DC101 targets the stromal vasculature in a stromal ries [progressive disease (PD) and other RECIST outcome] vessel phenotype model, which leads to a reduction in during patient treatment periods (for this dataset, best the stromal compartment but with negligible effect on RECIST responses also had longest duration). The proba- tumor size bility of the stromal vessel phenotype group having a poorer Tumor-embedded vessels were not detected in the Calu-3 response (PD category) to bevacizumab and FOLFIRI than model; however, vasculature was detected at a high density in that of the tumor vessel phenotype group was statisti- the stroma (Fig. 5A and B). Stromal vasculature was reduced cally significant [P ¼ 0.0487; 95% confidence interval (CI), by >50% in response to DC101 with the remaining stromal 1.24–¥]. Although it is not possible to conclude that the vessels exhibited a greater mean vessel area than those in differential response is a direct result of the treatment with untreated Calu-3 tumors (Fig. 5A). Qualitative assessment bevacizumab alone, these data suggest that the tumor vessel of CD31-a-SMA IF-stained tumors revealed the proportion phenotype appears to be more sensitive to the combination of pericyte-covered stromal vessels increased in treated tu- therapy than the stromal vessel phenotype. mors (Fig. 5B). These data indicate that while treatment with DC101 reduces small immature vessels, it fails to reduce the larger more mature vasculature within the stroma. Discussion To examine the influence of the stromal vessel pheno- This study suggests that human tumor types can be type on the response to DC101, we conducted a detailed broadly categorized according to the spatial distribution biomorphometric analysis of DC101-treated Calu-3 of their blood vessels in relation to the tumor cells and xenograft tumors. In Calu-3 tumors, DC101 treatment other stromal components. Across a broad panel of human led to reduction in the area of the stromal compartment tumor types, independent of disease type, tumors largely and an increase in the area occupied by tumor cells with adopted either a tumor vessel or stromal vessel phenotype. negligible effect on tumor size (Fig. 5C). Analysis of the The tumor vessel phenotype appears to be indicative of stromal compartment for myofibroblasts using a-SMA tumors that are likely to show greater single-agent and PDGFRb staining revealed a significant reduction in responses to anti-angiogenic drugs targeting the VEGF stromal myofibroblast levels (Fig. 5D). DC101 treatment signaling axis. Although we were able to separate different also led to a significant increase in the mean diameter of tumor types using this approach, no tumor type studied tumor nests (from 200 to 450 mm, P ¼ 0.04); however, was exclusively one type of architecture. For example, effects on the proliferative index and density of the tumor although RCC tumors are predominantly of the tumor were negligible at this time point (Supplementary Fig. S4a vessel phenotype, some exhibit a stromal vessel architec- and S4c). The recruitment of inflammatory cells was also ture. Conversely, while CRC tumors are dominated by the assessed using F4/80 (macrophages) and Gr-1 (neutro- stromal vessel phenotype, a subset exhibits the tumor phils). DC101 treatment resulted in a small but signifi- vessel architecture. Heterogeneity is also evident within cant increase in the tumor macrophage content and a individual tumors. In particular, tumor types with a stro- significant accumulation of Gr-1–positive cells in the mal vessel architecture have regions with tumor vessel stroma (Supplementary Fig. S4b and S4c). In conclusion, architecture, which may represent more angiogenic regions the reduction of stromal angiogenic vasculature by of the tumor. Tumor vessel vasculature is embedded within DC101 in the Calu-3 model had the greatest impact on the tumor cell mass. This forms a major structural feature the architecture and cellular composition of the stroma, of the tumor, facilitating intratumoral blood flow. In rather than the tumor compartment, but with a negligible contrast, in stromal vessel tumors, the vascular supply effect on the overall tumor mass. develops almost exclusively in the stromal compartment, which separates the tumor cells from the vessels, poten- The two phenotypes correlate with different clinical tially creating more mature vessels. efficacies to bevacizumab and FOLFIRI in metastatic The tumor vessel phenotype was associated with differ- CRC ences in the distribution of other cells, specifically the The relationship between phenotype and RECIST presence of high levels of macrophages. In stromal vessel response to a combination of VEGFi and oxaliplatin-based tumors, these macrophages did not commonly associate chemotherapy in patients with metastatic CRCs was inves- with the tumor cells perhaps indicating a different role for tigated. Two TMAs consisting of surgical tumor samples the macrophage in these tumors. A possible explanation for from 56 patients with metastatic CRC (Tristar) were chro- these observations is that the intratumoral vasculature mogenically stained for CD31-a-SMA and scored for the observed in the tumor vessel phenotype, but absent from

www.aacrjournals.org Clin Cancer Res; 19(24) December 15, 2013 OF9

Smith et al.

A Calu-3 (1) B Control DC101 treated ) 2 700 600 500 P = 0.0001 400 300 200 CD31 Stromal MVD 100 0 (number vessels per mm

μ Control 50 μm 50 m

DC101 treated 200 180 P = 0.04 160 140 Stromal MVA

120 -SMA

100 α per vessel) 80 2

m 60 μ (

40 CD31/ 20 0 Stromal microvessel area 100 μm 100 μm

Control

DC101 treated C Calu-3 (1) Calu-3 (1) Control DC101 treated

70 P = 0.09 60 50 P = 0.001 40 P = 0.001 30 Haematoxylin 200 μm μ 20 P = 0.35 200 m 10

Relative proportions of 0 xenograft compartments (%)

Control

DC101 treated overlay Compartment Tumour 200 μm 200 μm Stroma Necrotic space

Calu-3 (1)

100 Control α-SMA 90 β 80 PDGFR μ 70 100 m 100 μm 60 50 40 P = 0.0005

% Positivity 30 20 P = 0.009 10 0 DC101 treated Control 100 μm 100 μm

DC101 treated

OF10 Clin Cancer Res; 19(24) December 15, 2013 Clinical Cancer Research

Tumor Stromal Phenotypes Deﬁne VEGF Sensitivity

the stromal vessel phenotype, would facilitate extravasation of immune infiltrate into the tumor. In contrast, in the 30 PD stromal vessel, phenotype macrophages would extravasate into the dense stroma surrounding the tumor nests which 25 SD may impair their migration into the tumor. On the basis of PR E-cadherin loss and vimentin gain, the tumor vessel phe- 20 CR notype was more associated with tumor cells of a mesen- n = 56) chymal phenotype. It is conceivable that the tissue of origin 15 or subsequent progression of tumor to a more mesenchymal phenotype could influence reciprocal signaling 10 between the tumor and the tumor microenvironment, thereby influencing the extent of stromal recruitment and number Patient ( its subsequent architecture. However, we do not currently 5 have a mechanism to explain this. While it remains challenging to acquire human tumor 0 samples with associated outcome data following treatment TV SV with VEGF signaling inhibitors, we were able to obtain Phenotype TMAs of samples from patients with CRCs treated with bevacizumab þ FOLFIRI. Although these samples were Figure 6. Correlation of phenotype with RECIST response to bevacizumab limited in number, and only RECIST response category and FOLFIRI in metastatic CRC. Data were derived from a CD31-a-SMA information was available, they did allow initial explora- immunostained TMA consisting of 56 metastatic CRC samples (n ¼ 2 tion of the hypothesis that tumor vessel and stromal vessel cores per tumor) from patients before treatment. Tumors were classified tumor s may respond differently to therapy. The analysis as either tumor vessel or stromal vessel phenotypes by 2 observers and from the TMA is valid as previous data on baseline samples compared to best RECIST outcome for the treatment period of each patient. CR, complete response; PR, partial response; SD, stable sets showed good concordance between the whole tumor disease. On the basis of the dataset, Fisher exact test showed that the sample and the TMA. The distribution of tumor vessel stromal vessel phenotype group has higher probability of having a poor versus stromal vessel was largely representative of the larger response (PD category) to bevacizumab and FOLFIRI than that of the fi P ¼ datasets for CRCs. Interestingly, in this dataset, the tumor tumor vessel phenotype group with statistical signi cance ( 0.0487; 95% CI, 1.24–¥). vessel phenotype was associated with good response, with no examples of PD associated with this phenotype, suggest- ing that the tumor vessel phenotype may be more sensitive establish whether the phenotypes are maintained during to the combination therapy. This supports the idea that the disease progression (e.g., metastatic lesions) and on pro- tumor vessel or stromal vessel should be considered when gression following therapeutic intervention to determine differentiating response of these tumors to angiogenic ther- the use of stratifying tumor types based on phenotype using apy. It would be interesting to understand in late-stage archival samples. CRCs, where VEGF inhibitors can be used as a single agent, Preclinical tumor xenografts have been used to define whether differential response or disease control is associ- many of the mechanisms with the potential to influence ated with the differences in phenotype. As progression-free response to VEGFi, other angiogenic factors (7), bone survival (PFS) and overall survival (OS) information was marrow–derived cells (5, 6), or by recruitment of stromal not available for these samples, further analyses to correlate fibroblasts/pericytes into the tumor which drive the matu- phenotypes with these survival outcomes are required to ration of tumor vessels (8–10). Our analysis suggests that understand the use of this approach to predict long-term those mechanisms of resistance defined in preclinical tumor outcome in diseases dominated by stromal vessel morphol- xenografts may be most relevant to those diseases that ogy. Only primary tumor samples were analyzed in this display a tumor vessel phenotype. It is probable that in study. Often archival diagnostic samples from the primary selecting tumor models that grow quickly to facilitate drug tumor are the only tissue available to evaluate the tumor testing, stromal-rich tumors have been severely underrep- biomarker status of a patient. It will be important to resented by virtue of their growth characteristics.

Figure 5. The stromal neovasculature is targeted by DC101 in a tumor model of the stromal vessel phenotype, Calu-3 which leads to stromal changes with þ þ þ negligible effect on tumor size. A, MVD (number of CD31 vessels per mm2) and microvessel area (MVA, total CD31 area (mm2)/total number of CD31 vessels) of Genie classified stromal compartment of IgG1 control and DC101-treated Calu-3 tumor s (growth curve study 1). B, chromogenic (CD31, DAB, brown) and fluorescent [CD31-AF488 (green), a-SMA -AF555 (red), and DAPI counterstain (blue)] immunostained control and DC101-treated Calu-3 tumor s (growth curve 1). For CD31 images, black or gray arrows point to small and large stromal vessels, respectively (S, stroma; T, tumor). For CD31-a-SMA images, white or gray arrows indicate naked and pericyte-covered vessels, respectively. C, Genie (Aperio Technologies Ltd) classification and quantification of tumor, stromal, and necrotic areas (% number of tumor, stromal, or necrotic pixels/total number of pixels) using digitally captured images of hematoxylin- stained sections for DC101 (15 mg/kg twice weekly, n ¼ 6) and IgG1 (control, n ¼ 6) treated arms of Calu-3 growth studies [1 and 2 (data not shown)]. Hematoxylin (blue) staining of control and DC101-treated Calu-3 tumors and overlay of Genie classified tumor, stromal, and necrotic compartments for Calu-3 þ þ growth curves 1 and 2 (data not shown). D, stromal myofibroblast content (percentage number of aSMA or PDGFRb pixels/total number pixels) within Genie classified regions of control and DC101-treated tumor s (growth curve 1). Mean SEM and P values are shown for all biomarkers.

www.aacrjournals.org Clin Cancer Res; 19(24) December 15, 2013 OF11

Smith et al.

Understanding the implications of the stromal vessel Our findings offer an alternative way to interpret the architecture on tumor cell function and response to drugs effects we have seen with VEGFi therapy in the clinic. If will be important. The Calu-3 model (stromal vessel tumors with either a tumor vessel or stromal vessel phenotype), which exhibited a slower growth rate and architecture are viewed as distinct from the point of view showed a poor response to the specific VEGFR2-blocking of tumor angiogenesis, then it also challenges whether we antibody DC101, expressed higher levels of human genes have interpreted current biomarkers appropriately. For thought to play a role in recruitment of stroma than example, while biomarker changes in vessel numbers or models representative of the tumor vessel phenotype. vascular function in response to VEGFi treatment could These genes, PDGF-A, C and D, FGF-2, interleukin (IL)- conceivably be observed in tumors with either a tumor 8 are known to influence fibroblast function as well as vessel or stromal vessel phenotype, it is possible that these resistance to VEGF signaling inhibitors (9–12, 27–30). The may only translate into therapeutic benefit in the former. high stromal content would generate a distinct microen- Given that predominantly stromal vessel diseases (e.g., vironment where the mature vessels in the stroma may be CRC) will still have a small proportion of tumors with a resistant to therapy, an effect observed with VEGF signal- tumor vessel phenotype, it would also be interesting to ing inhibitors in preclinical tumor xenografts (3, 11), the determine whether differences in response are seen with- stromal phenotype may be a more extreme representation in these tumor subsets. This segmentation of disease may of this resistance mechanism. Interestingly, Calu-3 tumor offer an approach to prioritize tumor types that are likely stroma had a high local vessel density that was reduced to show a clinical response to VEGFi therapy and poten- following DC101 treatment in this model, but a significant tially other vascular directed agents. We suggest that number of established mature residual vessels remained. diseases that predominantly display a tumor vessel phe- In addition, although treatment with DC101 led to a notype would show good single-agent response to such decrease in the density of the stroma, the general archi- agents. tecture of the stroma was maintained. This suggests that in This study has a number of limitations. Although the Calu-3 tumors, there may be a role for the vasculature in preclinical evaluation of the concept uses several tumor maintaining the stroma or that changes in the tumor vessel models treated with DC101, because of the scarcity of compartment following a reduction in vessels lead to a tumor xenografts that represent the stromal vessel pheno- reduction in reactive stroma. Treating human stromal type, we were limited to one stromal vessel model (Calu-3). vessel tumors with VEGFi may change the apparent vessel Studies using additional VEGR antagonists at different permeability, perfusion or blood flow as determined by exposures in models that represent a range of phenotypes, imaging or reduces vessels as measured by other biomarker including additional stromal vessel phenotypes and models approaches, but without reaching a threshold that is heterogeneous for both phenotypes, would be important in sufficient to impact upon tumor cell growth or survival. investigating this concept further. In addition, a more In stromal vessel tumors, pruning of immature vasculature comprehensive evaluation of the phenotypes in specific and retention of larger mature vessels in the stroma by tumor types is required to better understand (i) prevalence VEGFi could alter interstitial pressure within the tumor and heterogeneity, (ii) prognostic and predictive value, and consistent with the vascular normalization hypothesis (iii) phenotype changes associated with disease progression (31) and influence drug delivery. Tumor types that are or following therapy. predominantly stromal vessel (e.g., CRC, NSCLC, BC) do The vascular architecture is not the only parameter that is not exhibit dramatic responses to VEGFi monotherapy in likely to determine the response of an individual tumor to the clinic but alterations in stromal density may give VEGFi therapy. Within the context of each phenotype, other benefit in combination or with single-agent therapy in factors may further influence the outcome following ther- late-stage disease. Although we have only conducted a apy. For example, with a vascular targeting therapy, it is limited analysis, it will be interesting to examine how reasonable to expect factors that determine the metabolic reductions in tumor vasculature influence tumor cell sur- status of the cells, or the degree of environmental stress the vival or growth in other models with a high stromal tumor cell can withstand, would also influence the likeli- content. Here, we focused on testing VEGF signaling hood of obtaining an objective tumor response. In partic- inhibition by using an antibody that specifically antago- ular, the stromal architecture may influence features such as nized VEGFR2 on the murine vasculature. Many small- the intrinsic metabolic status of the tumor cells either molecule VEGFR inhibitors have also been developed, but directly, or by creating a dependency on metabolic coupling these compounds have broader pharmacology profiles between the stromal fibroblast and tumor to promote and inhibit additional kinase such as PDGFR and related anabolic growth (the "reverse Warburg effect"; ref. 32). It family members which impart additional effects in the is unclear what features of the tumor cell determine the stroma or tumor. It will be of future interest to also model vessel phenotypes, but a better understanding of the under- such small-molecule inhibitors preclinically taking into lying mechanisms could help further refine patient selection account their clinical exposure profiles, to determine strategies for VEGFi therapy. whether any differential effects are delivered with these In conclusion, we suggest that consideration of the tumor mixed pharmacology agents on tumor vessel and stromal stromal architecture may be an important determinant of vessel tumors. whether tumors will be therapeutically susceptible to

OF12 Clin Cancer Res; 19(24) December 15, 2013 Clinical Cancer Research

Tumor Stromal Phenotypes Deﬁne VEGF Sensitivity

treatment with VEGFi monotherapy and potentially other Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N.R. Smith, M. Farren, A.J.C. Pommier, R. Swann, vascular modulating agents. The ability to stratify VEGFi K. McDaid, J. Kendrew, C. Womack responses in preclinical models and human disease on the Analysis and interpretation of data (e.g., statistical analysis, biosta- basis of a tumor vessel or stromal vessel phenotype warrants tistics, computational analysis): N.R. Smith, M. Farren, A.J.C. Pommier, X. Wang, S. Mistry, J. Kendrew, C. Womack, S. Wedge, S.T. Barry wider evaluation. Writing, review, and/or revision of the manuscript: N.R. Smith, A.J.C. Pommier, J. Kendrew, C. Womack, S. Wedge, S.T. Barry Disclosure of Potential Conﬂicts of Interest Administrative, technical, or material support (i.e., reporting or orga- nizing data, constructing databases): N.R. Smith, D. Baker, A.J.C. Pom- All authors are current or former AstraZeneca employees and share- mier, K. McDaid, S.T. Barry holders. M. Farren was employed as a Post Doc and S. Wedge as Senior Study supervision: N.R. Smith, S.T. Barry Principal Scientist. J. Kendrew and S.T. Barry have Ownership Interest The costs of publication of this article were defrayed in part by the (including patents) as AstraZeneca Shareholder. No potential conﬂicts of payment of page charges. This article must therefore be hereby marked interest were disclosed. advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. Authors' Contributions Conception and design: N.R. Smith, J. Kendrew, S.T. Barry Development of methodology: N.R. Smith, M. Farren, A.J.C. Pommier, Received June 17, 2013; revised August 28, 2013; accepted September 4, S. Mistry, S.T. Barry 2013; published OnlineFirst September 12, 2013.

References 1. Crawford Y, Ferrara N. Tumor and stromal pathways mediating refrac- endothelial growth factor receptor tyrosine kinase inhibitor, in patients toriness/resistance to anti-angiogenic therapies. Trends Pharmacol with recurrent glioblastoma. J Clin Oncol 2010;28:2817–23. Sci 2009:30:624–30. 16. Friedman HS, Prados MD, Wen PY, Mikkelsen T, Schiff D, Abrey LE, 2. Sitohy B, Nagy JA, Dvorak HF. Anti-VEGF/VEGFR therapy for cancer: et al. Bevacizumab alone and in combination with irinotecan in recur- reassessing the target. Cancer Res 2012;72:1909–14. rent glioblastoma. J Clin Oncol 2009;27:4733–40. 3. Bergers G, Hanahan D. Modes of resistance to anti-angiogenic ther- 17. Winder T, Lenz HJ. Vascular endothelial growth factor and epidermal apy. Nat Rev Cancer 2008;8:592–603. growth factor signaling pathways as therapeutic targets for colorectal 4. Ivy SP, Wick JY, Kaufman BM. An overview of small-molecule inhibi- cancer. Gastroenterology 2010;138:2163–76. tors of VEGFR signaling. Nat Rev Clin Oncol 2009;6:569–79. 18. Das M, Wakelee H. Targeting VEGF in lung cancer. Expert Opin Ther 5. Muramatsu M, Yamamoto S, Osawa T, Shibuya M. Vascular endo- Targets 2012;16:395–406. thelial growth factor receptor-1 signaling promotes mobilization of 19. Lam ET, Ringel MD, Kloos RT, Prior TW, Knopp MV, Liang J, et al. macrophage lineage cells from bone marrow and stimulates solid Phase II clinical trial of sorafenib in metastatic medullary thyroid tumor growth. Cancer Res 2010;70:8211–21. cancer. J Clin Oncol 2010;28:2323–30. 6. Shojaei F, Wu X, Qu X, Kowanetz M, Yu L, Tan M, et al. G-CSF- 20. Bible KC, Suman VJ, Molina JR, Smallridge RC, Maples WJ, Menefee initiated myeloid cell mobilization and angiogenesis mediate tumor ME, et al. Efﬁcacy of pazopanib in progressive, radioiodine-refractory, refractoriness to anti-VEGF therapy in mouse models. PNAS 2009; metastatic differentiated thyroid cancers: results of a phase 2 consor- 106:6742–7. tium study.; Endocrine Malignancies Disease Oriented Group; Mayo 7. Welti JC, Gourlaouen M, Powles T, Kudahetti SC, Wilson P, Berney Clinic Cancer Center; Mayo Phase 2 Consortium. Lancet Oncol DM, et al. Fibroblast growth factor 2 regulates endothelial cell sensi- 2010;11:962–72. tivity to sunitinib. Oncogene 2011;30:1183–93. 21. Antonarakis ES, Carducci MA. Targeting angiogenesis for the 8. Cascone T, Herynk MH, Xu L, Du Z, Kadara H, Nilsson MB, et al. treatment of prostate cancer. Expert Opin Ther Targets 2012;16: Upregulated stromal EGFR and vascular remodeling in mouse xeno- 365–76. graft models of angiogenesis inhibitor-resistant human lung adeno- 22. Mackey JR, Kerbel RS, Gelmon KA, McLeod DM, Chia SK, Rayson D, carcinoma. J Clin Invest 2011;121:1313–28. et al. Controlling angiogenesis in breast cancer: a systematic review of 9. Kitadai Y, Sasaki T, Kuwai T, Nakamura T, Bucana CD, Fidler IJ, et al. anti-angiogenic trials. Cancer Treat Rev 2012;38:673–88. Targeting the expression of platelet-derived growth factor receptor by 23. Raymond E, Dahan L, Raoul JL, Bang YJ, Borbath I, Lombard-Bohas reactive stroma inhibits growth and metastasis of human colon car- C, et al. Sunitinib malate for the treatment of pancreatic neuroendo- cinoma. Am J Pathol 2006;169:2054–65. crine tumors. N Engl J Med 2011;364:501–13. 10. Pietras K, Pahler J, Bergers G, Hanahan D. Functions of paracrine 24. Kindler HL, Wroblewski K, Wallace JA, Hall MJ, Locker G, Nattam S, PDGF signaling in the pro-angiogenic tumor stroma revealed by et al. Gemcitabine plus bevacizumab compared with gemcitabine plus pharmacological targeting. PLoS Med 2008;5:e19. placebo in patients with advanced pancreatic cancer: phase III trial of 11. Crawford Y, Kasman I, Yu L, Zhong C, Wu X, Modrusan Z, et al. PDGF- the Cancer and Leukemia Group B (CALGB 80303). J Clin Oncol C mediates the angiogenic and tumorigenic properties of ﬁbroblasts 2010;28:3617–22. associated with tumors refractory to anti-VEGF treatment. Cancer Cell 25. Kindler HL, Ioka T, Richel DJ, Bennouna J, Letourneau R, Okusaka 2009;15:21–34. T, et al. Axitinib plus gemcitabine versus placebo plus gemcita- 12. Farren M, Weston S, Brown H, Broadbent N, Powell S, Shaw R, et al. bine in patients with advanced pancreatic adenocarcinoma: a Expression of stromal genes associated with the angiogenic response double-blind randomised phase 3 study. Lancet Oncol 2011;12: are not differentiated between human tumour xenografts with diver- 256–62. gent vascular morphologies. Angiogenesis 2012;15:555–68. 26. O'Reilly EM, Niedzwiecki D, Hall M, Hollis D, Bekaii-Saab T, Pluard T, 13. Smith NR, Baker D, James NH, Ratcliffe K, Jenkins M, Ashton SE, et al. et al. A Cancer and Leukemia Group B phase II study of sunitinib malate Vascular endothelial growth factor receptors VEGFR-2 and VEGFR-3 in patients with previously treated metastatic pancreatic adenocarci- are localized primarily to the vasculature in human primary solid noma (CALGB 80603). Oncologist 2010;15:1310–9. cancers. Clin Cancer Res 2010;16:3548–61. 27. Prewett M, Huber J, Li Y, Santiago A, O'Connor W, King K, et al. 14. Sternberg CN, Davis ID, Mardiak J, Szczylik C, Lee E, Wagstaff J, et al. Antivascular endothelial growth factor receptor (fetal liver kinase 1) Pazopanib in locally advanced or metastatic renal cell carcinoma: monoclonal antibody inhibits tumor angiogenesis and growth of sev- results of a randomized phase III trial. J Clin Oncol 2010;28:1061–8. eral mouse and human tumors. Cancer Res 1999;59:5209–18. 15. Batchelor TT, Duda DG, di Tomaso E, Ancukiewicz M, Plotkin SR, 28. Wedge SR, Kendrew J, Hennequin LF, Valentine PJ, Barry ST, Brave Gerstner E, et al. Phase II study of cediranib, an oral pan-vascular SR, et al. AZD2171: a highly potent, orally bioavailable, vascular

www.aacrjournals.org Clin Cancer Res; 19(24) December 15, 2013 OF13

Smith et al.

endothelial growth factor receptor-2 tyrosine kinase inhibitor for the 31. Chauhan VP, Stylianopoulos T, Martin JD, Popovic Z, Chen O, Kamoun treatment of cancer. Cancer Res 2005;65:4389–400. WS, et al. Normalization of tumour blood vessels improves the delivery 29. Ebos JM, Lee CR, Christensen JG, Mutsaers AJ, Kerbel RS. Multiple of nanomedicines in a size-dependent manner. Nat Nanotechnol circulating proangiogenic factors induced by sunitinib malate are 2012;7:383–8. tumor-independent and correlate with antitumor efﬁcacy. Proc Natl 32. Whitaker-Menezes D, Martinez-Outschoorn UE, Lin Z, Ertel A, Acad Sci U S A 2007;104:17069–74. Flomenberg N, Witkiewicz AK, et al. Evidence for a stromal-epithe- 30. Casanovas O, Hicklin D, Bergers G, Hanahan D. Drug resistance by lial "lactate shuttle" in human tumors: MCT4 is a marker of evasion of antiangiogenic targeting of VEGF signaling in late stage oxidative stress in cancer-associated ﬁbroblasts. Cell Cycle 2011; pancreatic islet tumors. Cancer Cell 2005;8:299–309. 10:1772–83.

OF14 Clin Cancer Res; 19(24) December 15, 2013 Clinical Cancer Research

Tumor Stromal Architecture Can Define the Intrinsic Tumor Response to VEGF-Targeted Therapy

Neil R. Smith, Dawn Baker, Matthew Farren, et al.

Clin Cancer Res Published OnlineFirst September 12, 2013.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-13-1637

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2013/09/13/1078-0432.CCR-13-1637.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/early/2013/12/12/1078-0432.CCR-13-1637. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.